xref: /openbmc/qemu/migration/migration.h (revision 53ba2eee)
1 /*
2  * QEMU live migration
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef QEMU_MIGRATION_H
15 #define QEMU_MIGRATION_H
16 
17 #include "exec/cpu-common.h"
18 #include "hw/qdev-core.h"
19 #include "qapi/qapi-types-migration.h"
20 #include "qemu/thread.h"
21 #include "qemu/coroutine_int.h"
22 #include "io/channel.h"
23 #include "net/announce.h"
24 #include "qom/object.h"
25 
26 struct PostcopyBlocktimeContext;
27 
28 #define  MIGRATION_RESUME_ACK_VALUE  (1)
29 
30 /*
31  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
32  * the benefit that all the chunks are 64 pages aligned then the
33  * bitmaps are always aligned to LONG.
34  */
35 #define CLEAR_BITMAP_SHIFT_MIN             6
36 /*
37  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
38  * default value to use if no one specified.
39  */
40 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
41 /*
42  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
43  * big enough and make sure we won't overflow easily.
44  */
45 #define CLEAR_BITMAP_SHIFT_MAX            31
46 
47 /* State for the incoming migration */
48 struct MigrationIncomingState {
49     QEMUFile *from_src_file;
50 
51     /*
52      * Free at the start of the main state load, set as the main thread finishes
53      * loading state.
54      */
55     QemuEvent main_thread_load_event;
56 
57     /* For network announces */
58     AnnounceTimer  announce_timer;
59 
60     size_t         largest_page_size;
61     bool           have_fault_thread;
62     QemuThread     fault_thread;
63     QemuSemaphore  fault_thread_sem;
64     /* Set this when we want the fault thread to quit */
65     bool           fault_thread_quit;
66 
67     bool           have_listen_thread;
68     QemuThread     listen_thread;
69     QemuSemaphore  listen_thread_sem;
70 
71     /* For the kernel to send us notifications */
72     int       userfault_fd;
73     /* To notify the fault_thread to wake, e.g., when need to quit */
74     int       userfault_event_fd;
75     QEMUFile *to_src_file;
76     QemuMutex rp_mutex;    /* We send replies from multiple threads */
77     /* RAMBlock of last request sent to source */
78     RAMBlock *last_rb;
79     void     *postcopy_tmp_page;
80     void     *postcopy_tmp_zero_page;
81     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
82     GArray   *postcopy_remote_fds;
83 
84     QEMUBH *bh;
85 
86     int state;
87 
88     bool have_colo_incoming_thread;
89     QemuThread colo_incoming_thread;
90     /* The coroutine we should enter (back) after failover */
91     Coroutine *migration_incoming_co;
92     QemuSemaphore colo_incoming_sem;
93 
94     /*
95      * PostcopyBlocktimeContext to keep information for postcopy
96      * live migration, to calculate vCPU block time
97      * */
98     struct PostcopyBlocktimeContext *blocktime_ctx;
99 
100     /* notify PAUSED postcopy incoming migrations to try to continue */
101     bool postcopy_recover_triggered;
102     QemuSemaphore postcopy_pause_sem_dst;
103     QemuSemaphore postcopy_pause_sem_fault;
104 
105     /* List of listening socket addresses  */
106     SocketAddressList *socket_address_list;
107 
108     /* A tree of pages that we requested to the source VM */
109     GTree *page_requested;
110     /* For debugging purpose only, but would be nice to keep */
111     int page_requested_count;
112     /*
113      * The mutex helps to maintain the requested pages that we sent to the
114      * source, IOW, to guarantee coherent between the page_requests tree and
115      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
116      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
117      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
118      * for that page already.  This is intended so that the mutex won't
119      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
120      * this should be enough to make sure the page_requested tree always
121      * contains valid information.
122      */
123     QemuMutex page_request_mutex;
124 };
125 
126 MigrationIncomingState *migration_incoming_get_current(void);
127 void migration_incoming_state_destroy(void);
128 /*
129  * Functions to work with blocktime context
130  */
131 void fill_destination_postcopy_migration_info(MigrationInfo *info);
132 
133 #define TYPE_MIGRATION "migration"
134 
135 typedef struct MigrationClass MigrationClass;
136 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
137                      MIGRATION_OBJ, TYPE_MIGRATION)
138 
139 struct MigrationClass {
140     /*< private >*/
141     DeviceClass parent_class;
142 };
143 
144 struct MigrationState {
145     /*< private >*/
146     DeviceState parent_obj;
147 
148     /*< public >*/
149     QemuThread thread;
150     QEMUBH *cleanup_bh;
151     QEMUFile *to_dst_file;
152     /*
153      * Protects to_dst_file pointer.  We need to make sure we won't
154      * yield or hang during the critical section, since this lock will
155      * be used in OOB command handler.
156      */
157     QemuMutex qemu_file_lock;
158 
159     /*
160      * Used to allow urgent requests to override rate limiting.
161      */
162     QemuSemaphore rate_limit_sem;
163 
164     /* pages already send at the beginning of current iteration */
165     uint64_t iteration_initial_pages;
166 
167     /* pages transferred per second */
168     double pages_per_second;
169 
170     /* bytes already send at the beginning of current iteration */
171     uint64_t iteration_initial_bytes;
172     /* time at the start of current iteration */
173     int64_t iteration_start_time;
174     /*
175      * The final stage happens when the remaining data is smaller than
176      * this threshold; it's calculated from the requested downtime and
177      * measured bandwidth
178      */
179     int64_t threshold_size;
180 
181     /* params from 'migrate-set-parameters' */
182     MigrationParameters parameters;
183 
184     int state;
185 
186     /* State related to return path */
187     struct {
188         QEMUFile     *from_dst_file;
189         QemuThread    rp_thread;
190         bool          error;
191         QemuSemaphore rp_sem;
192     } rp_state;
193 
194     double mbps;
195     /* Timestamp when recent migration starts (ms) */
196     int64_t start_time;
197     /* Total time used by latest migration (ms) */
198     int64_t total_time;
199     /* Timestamp when VM is down (ms) to migrate the last stuff */
200     int64_t downtime_start;
201     int64_t downtime;
202     int64_t expected_downtime;
203     bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
204     int64_t setup_time;
205     /*
206      * Whether guest was running when we enter the completion stage.
207      * If migration is interrupted by any reason, we need to continue
208      * running the guest on source.
209      */
210     bool vm_was_running;
211 
212     /* Flag set once the migration has been asked to enter postcopy */
213     bool start_postcopy;
214     /* Flag set after postcopy has sent the device state */
215     bool postcopy_after_devices;
216 
217     /* Flag set once the migration thread is running (and needs joining) */
218     bool migration_thread_running;
219 
220     /* Flag set once the migration thread called bdrv_inactivate_all */
221     bool block_inactive;
222 
223     /* Migration is waiting for guest to unplug device */
224     QemuSemaphore wait_unplug_sem;
225 
226     /* Migration is paused due to pause-before-switchover */
227     QemuSemaphore pause_sem;
228 
229     /* The semaphore is used to notify COLO thread that failover is finished */
230     QemuSemaphore colo_exit_sem;
231 
232     /* The event is used to notify COLO thread to do checkpoint */
233     QemuEvent colo_checkpoint_event;
234     int64_t colo_checkpoint_time;
235     QEMUTimer *colo_delay_timer;
236 
237     /* The first error that has occurred.
238        We used the mutex to be able to return the 1st error message */
239     Error *error;
240     /* mutex to protect errp */
241     QemuMutex error_mutex;
242 
243     /* Do we have to clean up -b/-i from old migrate parameters */
244     /* This feature is deprecated and will be removed */
245     bool must_remove_block_options;
246 
247     /*
248      * Global switch on whether we need to store the global state
249      * during migration.
250      */
251     bool store_global_state;
252 
253     /* Whether we send QEMU_VM_CONFIGURATION during migration */
254     bool send_configuration;
255     /* Whether we send section footer during migration */
256     bool send_section_footer;
257 
258     /* Needed by postcopy-pause state */
259     QemuSemaphore postcopy_pause_sem;
260     QemuSemaphore postcopy_pause_rp_sem;
261     /*
262      * Whether we abort the migration if decompression errors are
263      * detected at the destination. It is left at false for qemu
264      * older than 3.0, since only newer qemu sends streams that
265      * do not trigger spurious decompression errors.
266      */
267     bool decompress_error_check;
268 
269     /*
270      * This decides the size of guest memory chunk that will be used
271      * to track dirty bitmap clearing.  The size of memory chunk will
272      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
273      * bitmap for each page to send (1<<0=1); N=10 means we will clear
274      * dirty bitmap only once for 1<<10=1K continuous guest pages
275      * (which is in 4M chunk).
276      */
277     uint8_t clear_bitmap_shift;
278 
279     /*
280      * This save hostname when out-going migration starts
281      */
282     char *hostname;
283 };
284 
285 void migrate_set_state(int *state, int old_state, int new_state);
286 
287 void migration_fd_process_incoming(QEMUFile *f, Error **errp);
288 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
289 void migration_incoming_process(void);
290 
291 bool  migration_has_all_channels(void);
292 
293 uint64_t migrate_max_downtime(void);
294 
295 void migrate_set_error(MigrationState *s, const Error *error);
296 void migrate_fd_error(MigrationState *s, const Error *error);
297 
298 void migrate_fd_connect(MigrationState *s, Error *error_in);
299 
300 bool migration_is_setup_or_active(int state);
301 bool migration_is_running(int state);
302 
303 void migrate_init(MigrationState *s);
304 bool migration_is_blocked(Error **errp);
305 /* True if outgoing migration has entered postcopy phase */
306 bool migration_in_postcopy(void);
307 MigrationState *migrate_get_current(void);
308 
309 bool migrate_postcopy(void);
310 
311 bool migrate_release_ram(void);
312 bool migrate_postcopy_ram(void);
313 bool migrate_zero_blocks(void);
314 bool migrate_dirty_bitmaps(void);
315 bool migrate_ignore_shared(void);
316 bool migrate_validate_uuid(void);
317 
318 bool migrate_auto_converge(void);
319 bool migrate_use_multifd(void);
320 bool migrate_pause_before_switchover(void);
321 int migrate_multifd_channels(void);
322 MultiFDCompression migrate_multifd_compression(void);
323 int migrate_multifd_zlib_level(void);
324 int migrate_multifd_zstd_level(void);
325 
326 int migrate_use_xbzrle(void);
327 int64_t migrate_xbzrle_cache_size(void);
328 bool migrate_colo_enabled(void);
329 
330 bool migrate_use_block(void);
331 bool migrate_use_block_incremental(void);
332 int migrate_max_cpu_throttle(void);
333 bool migrate_use_return_path(void);
334 
335 uint64_t ram_get_total_transferred_pages(void);
336 
337 bool migrate_use_compression(void);
338 int migrate_compress_level(void);
339 int migrate_compress_threads(void);
340 int migrate_compress_wait_thread(void);
341 int migrate_decompress_threads(void);
342 bool migrate_use_events(void);
343 bool migrate_postcopy_blocktime(void);
344 
345 /* Sending on the return path - generic and then for each message type */
346 void migrate_send_rp_shut(MigrationIncomingState *mis,
347                           uint32_t value);
348 void migrate_send_rp_pong(MigrationIncomingState *mis,
349                           uint32_t value);
350 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
351                               ram_addr_t start, uint64_t haddr);
352 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
353                                       RAMBlock *rb, ram_addr_t start);
354 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
355                                  char *block_name);
356 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
357 
358 void dirty_bitmap_mig_before_vm_start(void);
359 void dirty_bitmap_mig_cancel_outgoing(void);
360 void dirty_bitmap_mig_cancel_incoming(void);
361 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
362                                       Error **errp);
363 
364 void migrate_add_address(SocketAddress *address);
365 
366 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
367 
368 #define qemu_ram_foreach_block \
369   #warning "Use foreach_not_ignored_block in migration code"
370 
371 void migration_make_urgent_request(void);
372 void migration_consume_urgent_request(void);
373 bool migration_rate_limit(void);
374 
375 #endif
376